[07:39:24] --- jaltman has left: Disconnected
[09:10:38] --- Russ has become available
[12:07:22] --- jaltman has become available
[13:02:43] --- jaltman has left: Disconnected
[13:22:40] --- deason has become available
[13:48:32] <kaduk@mit.edu/barnowl> During buildworld, on the console:
FlushVCache refcount too large
afs_vop_reclaim: afs_FlushVCache failed code 16 vnode
vc 0xffffff80010ca8c8 vp 0xffffff006cdce000 tag afs, fid:
113.536890486.82842.46143, opens 0, writers 0
  states statd
A while later, buildworld gives:
===> lib/libarchive (all)
make: don't know how to make
/afs/sipb.mit.edu/project/freebsd/build/src/lib/libarchive/archive_write_set_compression_xz.c.
Stop
On a different machine:
athena% fs getfid
/afs/sipb.mit.edu/project/freebsd/build/src/lib/libarchive/archive_write_set_compression_xz.c
File
/afs/sipb.mit.edu/project/freebsd/build/src/lib/libarchive/archive_write_set_compression_xz.c
(536890486.82842.46143) contained in volume 536890486

So I guess I get to audit refcounting ... I'm kind of curious as to
where a bug would have to be in order to be this hard to tickle.  A
race condition, perhaps?
[15:23:23] <deason> something not locking the right thing when dec'ing that refcount may be very hard to reproduce... you only have a window of a few instructions
[15:24:18] <kaduk@mit.edu/barnowl> I.e. a race condition.
[15:27:03] <shadow@gmail.com/owl156D22D0> well, you can assert that the right thing is locked but it may require
wrapping the decrement
[15:52:28] <deason> well, it could also just be a lack of a dec in some case you hit rarely or something... possibly a vfs path only hit when contesting something a lot
[15:52:47] <deason> but yeah, a simple race would be the first thought
[16:18:50] --- jaltman has become available
[17:39:53] <kaduk@mit.edu/barnowl> cc -O2 -pipe -ggdb3 -fno-strict-aliasing -c y.tab.c -o error_table.o
error_table.y:2:23: error: afsconfig.h: No such file or directory
error_table.y:3:23: error: afs/param.h: No such file or directory
In file included from error_table.y:23:
/usr/include/sys/timeb.h:42:2: warning: #warning "this file includes
<sys/timeb.h> which is deprecated"
[18:02:22] <kaduk@mit.edu/barnowl> But make claims that it's using AFS_LDRULE, which has $(AFS_CFLAGS)
which has $(LWP_CFLAGS) which has $(COMMON_CFLAGS) which has
${COMMON_INCL} which has -I${TOP_OBJDIR}/src/config which should be
enough to pull in afsconfig, at least.
[19:03:38] <Russ> Are you sure you're not getting an implicit make rule to generate an object file from *.y source that bypasses the .c.o make rule?
[19:12:54] <kaduk@mit.edu/barnowl> No.
[19:17:38] <kaduk@mit.edu/barnowl> I posted a link to debugging make output in my mail.
[19:19:55] <kaduk@mit.edu/barnowl> Looks like
Examining error_table.y...modified 20:18:27 Oct 24, 2010...up-to-date.
Examining error_table.o...non-existent...modified before source (et_lex.lex.c)..
.out-of-date.
error_table.o:> = et_lex.lex.c
error_table.o:? = et_lex.lex.c
error_table.o:> = et_lex.lex.c error_table.y
error_table.o:? = et_lex.lex.c error_table.y
byacc -d error_table.y
cc -O2 -pipe -ggdb3 -fno-strict-aliasing -c y.tab.c -o error_table.o
Hm, is y.tab.c the default output of byacc?
[19:20:09] <Russ> Yup.
[19:21:27] <kaduk@mit.edu/barnowl> "So now what?"
[19:51:06] <shadow@gmail.com/owl156D22D0> we add a hardcoded rule back or add a .y.o rule, i guess?
[21:28:24] --- Russ has left: Disconnected
[22:12:13] --- deason has left
[23:19:20] --- canehan has become available
[23:37:37] --- reuteras has become available